Over the past six months, the concept of AI phones has transformed into a consensus for action across the industry.

At the recently concluded WAIC, companies such as MiniMax, StepFun, ZTE Nubia, Honor, and China Telecom Skywing showcased their AI phones. What truly matters isn't just their announcements but their actions. The roster of exhibitors often speaks louder than analytical reports in determining the viability of a particular sector.

Behind this surge of enthusiasm lies a clear policy signal. On July 12, the '15th Five-Year Plan for Expanding Consumption' was approved, explicitly stating the need to 'expand the consumption of digital products and enhance the effective supply of new-generation smart terminal products, including AI-enabled phones and computers, smart wearables, smart robots, and desktop-level 3D printing equipment'.

The inclusion of AI phones in this specialized plan carries significant weight.

Three days later, on July 15, regulatory compliance procedures were finalized. An official announcement listed the Record Filing Information for generative AI services on mobile devices, including Apple Intelligence, Huawei Xiaoyi AI, OPPO AndesGPT, vivo BlueHeart, Xiaomi HyperOS AI, Nubia Doubao Phone Model, and Samsung Galaxy AI.

Among these, Nubia's Doubao Phone Model stands out as unique—it is the only product deeply integrated with both an internet giant and a phone manufacturer, and the only one yet to launch on a large scale but already causing ripples in the industry.

The buzz from the release of Doubao AI Phone's engineering prototype last December still lingers, with a second-generation model now on the horizon.

As the baton passes from one generation to the next, a key technical shift emerges: the first-gen M153 relied on UI Automation (simulated clicks), allowing AI to operate apps through finger-like touch controls. However, it faced repeated blockades from mainstream apps like WeChat, Taobao, and Meituan. The second-gen model shifts to protocol interoperability, with Doubao now communicating directly with apps via protocols like MCP and A2A, instead of mimicking user taps.

According to supply chain sources, ByteDance has secured common permissions with partners in ride-hailing, food delivery, and ticketing services.

The first-gen's simulated click approach maintained a fragile peace amid data silos. The aggressive pivot suggests ByteDance is now serious about Doubao Phone. Yet, the equilibrium formed in the mobile internet era persists. Before diving into a complex strategic game, stakeholders weigh more than just the benefits.

I. A Complex Game of Strategies

While the first-gen Doubao Phone remained on the periphery of ByteDance's AI ambitions, its second-gen successor now stands at the core.

Visibly, Doubao Phone has been propelled to this stage by powerful external forces.

On December 1, 2025, the Doubao team launched a technical preview of Doubao Phone Assistant, debuting on the ZTE-collaborated engineering prototype Nubia M153, priced at 3,499 RMB with an initial stock of just around 30,000 units. The release carried a distinct technical validation tone, with restrained product design and inventory—avoiding bold claims like 'AI-native' or 'interaction revolution' common at AI hardware launches, and eschewing large-scale promotion.

This understated launch yielded surprising results. ByteDance likely didn't anticipate the low-key, rough engineering prototype selling out on launch day. But before the team could savor the surprise, backlash poured in from all sides.

On December 3, just days after launch, M153 users encountered WeChat warnings about 'abnormal login environments,' forcing accounts offline. Financial apps like Agricultural Bank of China soon followed with alerts. The Doubao team announced the voluntary suspension of WeChat and other top apps' operational capabilities. Subsequently, 8 out of 23 mainstream apps imposed restrictions: Taobao, Meituan, Alipay, Pinduoduo, and Maoyan blocked logins; WeChat, Gaode, and Damai allowed logins but rejected AI-driven operations.

The pressure peaked on January 26, 2026, when Tencent Chairman Pony Ma criticized Doubao Phone at an internal meeting: 'Using screen recording to transmit user phone and computer screens to the cloud is highly insecure and irresponsible. Tencent firmly opposes this.'

At that crossroads, ByteDance faced two paths: retreat or adapt differently. It chose the latter, confronting ecosystem barriers head-on.

Thus, the second-gen pivot became decisive. Reports indicate the new Nubia NaviX Ultra abandoned simulated clicks for MCP protocol-based app interoperability, accessing only super-apps that voluntarily offer MCP services and permissions. ByteDance began collaborating openly with former rivals.

Then, policy winds shifted.

On July 13, 2026, the '15th Five-Year Plan for Expanding Consumption' received official approval, explicitly listing AI phones as part of the new-generation smart terminal supply. Two days later, on July 15, the official filing list for on-device AI services included Doubao Phone's model. Compliance and policy green lights followed within a week.

Meanwhile, supply chain attitudes evolved. At its March 30 earnings call, ZTE—previously reserved—announced deepening cooperation with ByteDance and other ecosystem partners to advance next-gen Doubao AI phones. At WAIC, ZTE dedicated most of its booth to Nubia phones. Supply chain sources suggest the second-gen Doubao Phone will debut with around 100,000 units in initial stock, rising to around 200,000 overall.

With policy approvals, compliance, and supply chain support, ByteDance approached app vendors with a sincere protocol proposal—a hard offer to refuse.

But refusal wasn't the only response. Currently, only ride-hailing, food delivery, and ticketing partners have granted common permissions; attitudes from super-apps outside ByteDance's ecosystem remain unclear.

This is unsurprising. In the mobile internet era, OS vendors built the stage while apps performed, creating a relative equilibrium.

Today, AI muddies the waters. Neither OS nor app vendors will stand idly by. Everyone at the table is drawn into a complex game of strategies.

II. The Hidden Battlefield of AI Phones

The impending complex game ensures Doubao Phone's path won't be smooth.

In the short term, app openness will directly define Doubao Phone's capabilities and, consequently, consumer willingness to pay.

More critically, AI phones inherently threaten app vendors' survival. Traditional internet apps monetize by capturing user attention and engagement—time spent generates ads, transactions, and data. If users grow accustomed to voice-driven task completion (e.g., ordering food, hailing rides, booking tickets), app visits and dwell time will plummet. Even with a gentle approach, Doubao Phone risks cutting off user access to apps.

To persuade app vendors to open permissions, Doubao Phone must prove it expands the market rather than redistributes existing users. Yet a paradox emerges: user growth depends on AI feature richness, which hinges on app openness, which in turn depends on Doubao Phone's user base. It's a chicken-and-egg dilemma—without users, ByteDance's bargaining leverage remains limited.

Ten years ago, WeChat Mini Programs navigated a similar path. When launched in 2017, WeChat faced a structural dilemma akin to Doubao's today: if everything could be done within one app, why would others exist?

App vendors initially resisted Mini Programs. However, WeChat's 1 billion+ MAU, sticky social graph, and integrated payment ecosystem made access irresistible—fear of exclusion prevailed. Mini Programs evolved into an OS-like ecosystem, accommodating existing apps while fostering independent development.

Doubao Phone's dilemma shares similarities but differs fundamentally. WeChat Mini Programs operated within mobile internet's rules, with WeChat holding the reins. Doubao Phone's fate, however, lies temporarily in others' hands.

With around 200,000 units in initial stock, Doubao Phone lacks the scale to pressure top and mid-tier apps into 'join or exit' scenarios. During this phase of limited functionality and user growth, negotiations will likely be slow and iterative.

As Doubao Phone navigates app ecosystem friction, it must also contend with pressure from underlying OS vendors.

Currently, ByteDance's Doubao Phone functions as an on-device Agent: OS vendors provide bottom-layer permissions, apps offer MCP interfaces, and Doubao orchestrates. Unlike StepFun, which built an Agent OS from scratch, ByteDance's approach caps its capabilities based on OS vendors' willingness to cede control.

The July 15 filing list featured prominent names: Apple, Huawei, OPPO, vivo, and Xiaomi. These vendors boast massive user bases, mature distribution channels, and system-level permissions. Their core challenge is enhancing on-device AI—not lacking foundational capabilities. Compared to Doubao-WeChat tensions, OS vendors hold greater leverage over app partners.

However, divergent strategies among OS vendors create breathing room for Doubao Phone.

Apple, at WWDC 2026, restructured Siri as a standalone app, integrating a Gemini model co-developed with Google into a three-tier architecture (on-device model, private computing, cloud large model). If Siri becomes a system-level Agent, it too will disrupt app interests. Apple's strength lies in its ability to bypass negotiations with app vendors.

Xiaomi's MicLaw took a different route, encapsulating system capabilities into 50+ services within its 'human-vehicle-home' ecosystem, avoiding direct confrontation with app vendors.

Varying threat levels among OS vendors will shape app vendors' counterstrategies. The resulting time windows may offer Doubao Phone opportunities.

Yet, perhaps the most unpredictable variable comes from AI startups like StepFun.

On July 13, ahead of WAIC, StepFun unveiled Step AOS, a native Agent OS, and the STEPX Neo smartphone. Designed around AI capabilities, users interact via natural language with Agents that handle app calls—bypassing app interfaces entirely.

Most concerning for ByteDance is StepFun's initial ecosystem partners: Meituan, WPS, CapCut, Ctrip, Gaode, Alipay, Baidu, Didi, JD.com, Weibo, and Baidu Maps—covering nearly all daily high-frequency entry points.

STEPX Neo's pricing and launch plans remain undisclosed, and the depth of partnerships is unclear. Yet it underscores a reality: for app vendors, hedging bets is low-cost, while committing solely to one partner is risky. As a model company without legacy baggage, StepFun presents a less threatening negotiation partner.

For ByteDance, its mobile internet-era assets are both leverage and burden. Nevertheless, Doubao Phone remains a battlefield ByteDance cannot abandon.

III. From Performing to Building the Stage: More Than Just Capability

To date, ByteDance executives have rarely addressed 'Doubao Phone' publicly.

However, at this year's annual all-hands meeting, CEO Liang Rubo designated 'Scaling New Heights' as the 2026 theme, explicitly defining 'heights' in the short term as Doubao and Dola (its overseas version) assistant apps.

In a way, Doubao Phone serves as a critical springboard for Doubao's ascent.

When Liang interpreted 'Scaling New Heights,' he revisited 50 years of tech shifts: industry peaks emerge every 15–20 years, with Microsoft seizing PCs, Google the Web, and Apple mobile. Today, the window for such climbs is shortening.

ByteDance, a disruptor born in mobile internet's heyday, captured massive traffic via Toutiao and Douyin but never secured entry points at key tech inflection points—absent in the PC era, and mobile entry points belonged to Apple and Android. ByteDance has always been a performer, not a stage builder.

AI represents ByteDance's first entry-level opportunity since its founding. Liang's focus on Doubao underscores its strategic importance.

QuestMobile data shows Doubao reached 382 million MAU by June 2026, dominating China's AI app landscape. Volcano Engine reported Doubao's large model exceeded 180 trillion daily Token calls. By scale, Doubao leads—yet user stickiness lags. QuestMobile notes Doubao users average around 143.7 monthly minutes and 76.7 sessions—just under 5 minutes daily and around 2 minutes per session. In contrast, Douyin users exceed 120 daily minutes; WeChat, even more. Doubao remains a tool used as needed, not a habit integral to daily life.

This isn't unique to Doubao but a shared challenge for AI products. Current hot categories—conversational assistants, coding, and office tools—lean heavily toward utility. Users engage briefly and switch easily when better options emerge. Tools win users but rarely their time.

To solidify its AI-era foothold, ByteDance needs Doubao but can't rely on it alone.

Phones offer the smoothest path. They let ByteDance secure a lower-level entry point, transforming Doubao from an opened app into an Agent that opens everything for users—wielding far greater agency than in the app era.

ByteDance's past hardware attempts—from Dali Smart Lamps to PICO—shared this goal. Though none succeeded, ByteDance's determination to evolve from app factory to stage builder persists.

Today, ByteDance's approach is more restrained: consolidating resources around Doubao and launching hardware cautiously, letting Doubao Phone incrementally infiltrate daily life as Doubao's carrier.

Now, ByteDance boasts ample talent, resources, and organizational execution. AI phones challenge the mobile internet's hard-won equilibrium, but attempts to alter its trajectory demand substantial investment. As a major beneficiary of the last era, ByteDance enters this new battlefield with both historical credentials and formidable deterrence—app vendors guard against it far more vigilantly than startups.

This may be something even more difficult than technology. The route switch from simulating clicks to protocol interoperability can be completed within half a year, but achieving trust involves more than that.